Re: [TLS] tls 1.3: renegotiation

[Martin Thomson <[e-mail]>]

On 28 July 2014 15:24, Andrei Popov <[e-mail]> wrote:
> Are you saying that because HTTP/2 designers chose to ban renegotiation, the customers no longer need to perform TLS client auth in the middle of a connection? Perhaps I'm missing this point.
>
>> And it's not that alternatives don't exist, it's more the case that we disagree on the viability aspects, I think.
> The alternatives would need to address all the combinations of TLS and HTTP versions (and I believe the alternatives proposed so far don't). Also, there should be some way to encapsulate the alternatives such that sites don't need to fork client auth code for different TLS + HTTP version combinations.

That's fairly straightfoward.  My strawman proposal is
protocol-version-agnostic.

> Next, it is not desirable to add latency compared with the existing solution. And if we can get this to work for all application protocols (not just HTTP), even better. Hopefully the result will prove to be worth the design and implementation effort:).

I'm less sensitive to latency in these cases.  With a 0-RTT handshake,
it could be that HTTP/2 on a new connection would be faster than what
currently occurs, since from the point of HelloRequest, the server
needs to wait for two RTT from the client before it can send data.
(If the request contains a body that doesn't fit into the first flight
congestion window, then I might concede that it is slower.)